Related Experiment Video
Updated: May 4, 2026
09:11
Use of MRI-ultrasound Fusion to Achieve Targeted Prostate Biopsy
Published on: April 9, 2019
22.6K
Evaluation of prostate segmentation algorithms for MRI: the PROMISE12 challenge
Geert Litjens1, Robert Toth2, Wendy van de Ven1
1Radboud University Nijmegen Medical Centre, The Netherlands.
Medical Image Analysis
|January 15, 2014
Summary
The Prostate MR Image Segmentation (PROMISE12) challenge evaluated algorithms for segmenting prostate MRI scans. Top algorithms, particularly active appearance models, showed strong performance, with one even outperforming human experts, though further improvements are possible.
Area of Science:
- Medical Imaging
- Computer Vision
- Radiology
Background:
- Prostate MRI segmentation is crucial for cancer diagnosis and treatment planning.
- Evaluating segmentation algorithms across diverse datasets (multi-center, multi-vendor, multi-protocol) is challenging due to variations in image appearance and artifacts.
- The Prostate MR Image Segmentation (PROMISE12) challenge was established to facilitate fair comparison of segmentation methods.
Purpose of the Study:
- To compare the performance and robustness of various prostate MRI segmentation algorithms.
- To establish a standardized evaluation framework for prostate segmentation techniques.
- To identify leading algorithms and methods in the field.
Main Methods:
- The PROMISE12 challenge included 100 prostate MR cases from four centers with varied scanner manufacturers, field strengths, and protocols.
- Eleven teams submitted algorithms employing diverse methods, including active appearance models, atlas registration, and level sets.
- Evaluation utilized combined boundary and volume metrics, benchmarked against human expert performance.
Main Results:
- Algorithms demonstrated a wide range of performance, with active appearance model-based approaches generally outperforming multi-atlas registration in accuracy and speed.
- The winning algorithms (Imorphics and ScrAutoProstate) achieved high scores (85.72 and 84.29) and were significantly better than others (p<0.05).
- One algorithm surpassed the performance of a second human observer, but algorithm combination suggested potential for further enhancement.
Conclusions:
- The PROMISE12 challenge provided valuable insights into the state-of-the-art in prostate MRI segmentation.
- Active appearance models show promise for accurate and efficient prostate segmentation.
- While significant progress has been made, optimal performance in prostate segmentation remains an area for continued research and development.
